DDT: Dynamical Selective Dropping Threshold for Reactive Congestion Control
Pages 12 - 17
Abstract
Traditional congestion control algorithms (CCAs) frequently struggle to manage microbursts, resulting in performance degradation. Although RoCEv2 (RDMA over Converged Ethernet version 2) employs Priority Flow Control (PFC) to establish lossless networks and enhance burst tolerance, it does not guarantee low latency or high throughput due to challenges such as deadlocks, head-of-line (HoL) blocking, and congestion spreading. Consequently, investigating methods to reduce the side-effects of PFC in lossless networks or improve performance in lossy networks without PFC is critical. We introduce DDT, a mechanism designed to prevent timeouts by categorizing packets based on the impact after loss. DDT implements the Important Packet First and Important Packet Sustainability principles to prioritize important packets, ensuring their reliable transmission by actively dropping unimportant packets if necessary. DDT dynamically adjusts its selective dropping threshold based on the change ratio of the number of flows at switches, optimizing burst tolerance and reducing tail latency for short incast flows while minimizing the impact on the completion times of background flows. DDT is compatible with existing reactive window-based transport protocols. We have integrated DDT into HPCC and have demonstrated through large-scale simulations that: (1) DDT with PFC can reduce tail latency by up to 83.2% and greatly reduce the side effects of PFC on network performance; (2) DDT without PFC can mitigate timeouts and reduce tail latency by up to 83.7%.
References
[1]
2005. High-Capacity StrataXGS® Trident II Ethernet Switch Series. https://www.broadcom.com/products/ethernet-connectivity/switching/strataxgs/bcm56850-series.
[2]
2024. Open source code of our work. https://github.com/kalsasdf/HiNA-DCN_INET
[3]
Mohammad Alizadeh and Tom Edsall. 2013. On the Data Path Performance of Leaf-Spine Datacenter Fabrics. In Proceedings of IEEE Annual Symposium on High-Performance Interconnects, HOTI. 71–74. https://doi.org/10.1109/HOTI.2013.23
[4]
Mohammad Alizadeh, Albert G. Greenberg, David A. Maltz, and et al. 2010. Data center tcp (dctcp). In Proceedings of ACM Conference on Special Interest Group on Data Communication, SIGCOMM. 63–74. https://doi.org/10.1145/1851182.1851192
[5]
Hamidreza Almasi, Rohan Vardekar, Balajee Vamanan, and et al. 2023. Protean: Adaptive Management of Shared-Memory in Datacenter Switches. In Proceedings of IEEE Conference on Computer Communications, INFOCOM. 1–10. https://doi.org/10.1109/INFOCOM53939.2023.10229046
[6]
Wei Bai, Shuihai Hu, Kai Chen, and et al. 2020. One More Config is Enough: Saving (DC)TCP for High-speed Extremely Shallow-buffered Datacenters. In Proceedings of IEEE Conference on Computer Communications, INFOCOM. 2007–2016. https://doi.org/10.1109/INFOCOM41043.2020.9155280
[7]
Luiz André Barroso, Jeffrey Dean, Urs Hölzle, and et al. 2003. Web Search for a Planet: The Google Cluster Architecture. The Magazine on Microprocessors and Microsystems, Micro. 23, 2 (2003), 22–28. https://doi.org/10.1109/MM.2003.1196112
[8]
Luiz André Barroso, Mike Marty, David A. Patterson, and et al. 2017. Attack of the killer microseconds. ACM Transactions on Computer Systems, Commun. 60, 4 (2017), 48–54. https://doi.org/10.1145/3015146
[9]
Abhijit K Choudhury and Ellen L Hahne. 1998. Dynamic queue length thresholds for shared-memory packet switches. IEEE/ACM Transactions on Networking, TON. 6, 2, 130–140. https://doi.org/10.1109/90.664262
[10]
Jeffrey Dean and Luiz André Barroso. 2013. The tail at scale. ACM Transactions on Computer Systems, Commun. 56, 2 (2013), 74–80. https://doi.org/10.1145/2408776.2408794
[11]
Yixiao Gao, Qiang Li, Lingbo Tang, and et al. 2021. When Cloud Storage Meets RDMA. In Proceedings of USENIX Symposium on Networked Systems Design and Implementation, NSDI. 519–533. https://www.usenix.org/conference/nsdi21/presentation/gao
[12]
Chuanxiong Guo, Haitao Wu, Zhong Deng, and et al. 2016. RDMA over Commodity Ethernet at Scale. In Proceedings of ACM Conference on Special Interest Group on Data Communication, SIGCOMM. 202–215. https://doi.org/10.1145/2934872.2934908
[13]
Chuanxiong Guo, Lihua Yuan, Dong Xiang, and et al. 2015. Pingmesh: A Large-Scale System for Data Center Network Latency Measurement and Analysis. In Proceedings of ACM Conference on Special Interest Group on Data Communication, SIGCOMM. 139–152. https://doi.org/10.1145/2785956.2787496
[14]
Mark Handley, Costin Raiciu, Alexandru Agache, and et al. 2017. Re-architecting datacenter networks and stacks for low latency and high performance. In Proceedings of ACM Conference on Special Interest Group on Data Communication, SIGCOMM. 29–42. https://doi.org/10.1145/3098822.3098825
[15]
Shuihai Hu, Gaoxiong Zeng, Wei Bai, and et al. 2022. Aeolus: A Building Block for Proactive Transport in Datacenter Networks. IEEE/ACM Transactions on Networking, TON. 30, 2, 542–556. https://doi.org/10.1109/TNET.2021.3119986
[16]
Shuihai Hu, Yibo Zhu, Peng Cheng, and et al. 2019. Tagger: Practical PFC Deadlock Prevention in Data Center Networks. IEEE/ACM Transactions on Networking, TON. 27, 2 (2019), 889–902. https://doi.org/10.1109/TNET.2019.2902875
[17]
Gautam Kumar, Nandita Dukkipati, Keon Jang, and et al. 2020. Swift: Delay is simple and effective for congestion control in the datacenter. In Proceedings of ACM Conference on Special Interest Group on Data Communication, SIGCOMM. 514–528. https://doi.org/10.1145/3387514.3406591
[18]
Yang Li, Rui Miao, Hong Liu, and et al. 2019. HPCC: High precision congestion control. In Proceedings of ACM Conference on Special Interest Group on Data Communication, SIGCOMM. 44–58. https://doi.org/10.1145/3341302.3342085
[19]
Hwijoon Lim, Wei Bai, Yibo Zhu, and et al. 2021. Towards timeout-less transport in commodity datacenter networks. In Proceedings of European Conference on Computer Systems, EuroSys. ACM, 33–48. https://doi.org/10.1145/3447786.3456227
[20]
Radhika Mittal, Vinh The Lam, Nandita Dukkipati, and et al. 2015. TIMELY: RTT-based Congestion Control for the Datacenter. In Proceedings of ACM Conference on Special Interest Group on Data Communication, SIGCOMM. 537–550. https://doi.org/10.1145/2785956.2787510
[21]
Radhika Mittal, Alexander Shpiner, Aurojit Panda, and et al. 2018. Revisiting network support for RDMA. In Proceedings of ACM Conference on Special Interest Group on Data Communication, SIGCOMM. 313–326. https://doi.org/10.1145/3230543.3230557
[22]
Ieee 802.1 qbb. 2010. priority-based flow control. http://www.ieee802.org/1/pages/802.1bb.html
[23]
Brent E. Stephens, Alan L. Cox, Ankit Singla, and et al. 2014. Practical DCB for improved data center networks. In Proceedings of IEEE Conference on Computer Communications, INFOCOM. 1824–1832. https://doi.org/10.1109/INFOCOM.2014.6848121
[24]
Gaoxiong Zeng, Li Chen, Bairen Yi, and et al. 2022. Cutting Tail Latency in Commodity Datacenters with Cloudburst. In Proceedings of IEEE Conference on Computer Communications, INFOCOM. 600–609. https://doi.org/10.1109/INFOCOM48880.2022.9796898
[25]
Qiao Zhang, Vincent Liu, Hongyi Zeng, and et al. 2017. High-resolution measurement of data center microbursts. In Proceedings of the 2017 Internet Measurement Conference, IMC. 78–85. https://doi.org/10.1145/3131365.3131375
[26]
Yiwen Zhang, Gautam Kumar, Nandita Dukkipati, and et al. 2022. Aequitas: admission control for performance-critical RPCs in datacenters. In Proceedings of ACM Conference on Special Interest Group on Data Communication, SIGCOMM. 1–18. https://doi.org/10.1145/3544216.3544271
[27]
Yiran Zhang, Qingkai Meng, Chaolei Hu, and et al. 2024. Revisiting Congestion Control for Lossless Ethernet. In Proceedings of USENIX Symposium on Networked Systems Design and Implementation, NSDI. 131–148. https://www.usenix.org/conference/nsdi24/presentation/zhang-yiran
[28]
Yiying Zhang and Steven Swanson. 2015. A study of application performance with non-volatile main memory. In Symposium on Mass Storage Systems and Technologies, MSST. 1–10. https://doi.org/10.1109/MSST.2015.7208275
[29]
Yibo Zhu, Haggai Eran, Daniel Firestone, and et al. 2015. Congestion Control for Large-Scale RDMA Deployments. In Proceedings of ACM Conference on Special Interest Group on Data Communication, SIGCOMM. 523–536. https://doi.org/10.1145/2785956.2787484
[30]
Yibo Zhu, Nanxi Kang, Jiaxin Cao, and et al. 2015. Packet-Level Telemetry in Large Datacenter Networks. In Proceedings of ACM Conference on Special Interest Group on Data Communication, SIGCOMM. 479–491. https://doi.org/10.1145/2785956.2787483
Recommendations
Augmenting Proactive Congestion Control with Aeolus
APNet '18: Proceedings of the 2nd Asia-Pacific Workshop on NetworkingRecently, proactive congestion control solutions have drawn great attention in the community. By explicitly scheduling data transmissions based on the availability of network bandwidth, proactive solutions offer a lossless, near-zero queueing network ...
Congestion Control for High-speed Extremely Shallow-buffered Datacenter Networks
APNet '17: Proceedings of the First Asia-Pacific Workshop on NetworkingThe link speed in datacenters is growing fast, from 1Gbps to 100Gbps. However, the buffer size of commodity switches increases slowly, thus significantly outpaced by the link speed. In such extremely shallow-buffered datacenter networks, prior TCP/ECN ...
Explicit multipath congestion control for data center networks
CoNEXT '13: Proceedings of the ninth ACM conference on Emerging networking experiments and technologiesThe vast majority of application traffic in modern data center networks (DCNs) can be classified into two categories: throughput-sensitive large flows and latency-sensitive small flows. These two types of flows have the conflicting requirements on link ...
Comments
Information & Contributors
Information
Published In
July 2024
261 pages
ISBN:9798400710117
DOI:10.1145/3674399
Copyright © 2024 Owner/Author.
This work is licensed under a Creative Commons Attribution International 4.0 License.
Publisher
Association for Computing Machinery
New York, NY, United States
Publication History
Published: 30 July 2024
Check for updates
Author Tags
Qualifiers
- Research-article
- Research
- Refereed limited
Conference
ACM-TURC '24
Contributors
Other Metrics
Bibliometrics & Citations
Bibliometrics
Article Metrics
- 0Total Citations
- 43Total Downloads
- Downloads (Last 12 months)43
- Downloads (Last 6 weeks)43
Reflects downloads up to 04 Sep 2024
Other Metrics
Citations
View Options
View options
View or Download as a PDF file.
PDFeReader
View online with eReader.
eReaderHTML Format
View this article in HTML Format.
HTML FormatGet Access
Login options
Check if you have access through your login credentials or your institution to get full access on this article.
Sign in